On Mon, Feb 27, 2017 at 1:00 PM, David Ahern <d...@cumulusnetworks.com> wrote:
> On 2/27/17 12:34 PM, Eric Dumazet wrote:
>> On Mon, 2017-02-27 at 11:07 -0800, Cong Wang wrote:
>>> Andrey reported a NULL pointer deref bug in ipv6_route_ioctl()
>>> -> ip6_route_del() -> __ip6_del_rt_siblings() code path. This is
>>> because ip6_null_entry is returned in this path since ip6_null_entry
>>> is kinda default for a ipv6 route table root node. Quote from
>>> David Ahern:
>>>
>>> ip6_null_entry is the root of all ipv6 fib tables making it integrated
>>> into the table ...
>>>
>>> We should ignore any attempt of trying to delete it, like we do in
>>> __ip6_del_rt() path and several others.
>>>
>>> Reported-by: Andrey Konovalov <andreyk...@google.com>
>>> Signed-off-by: Cong Wang <xiyou.wangc...@gmail.com>
>>> ---
>>> net/ipv6/route.c | 7 +++++--
>>> 1 file changed, 5 insertions(+), 2 deletions(-)
>>>
>>> diff --git a/net/ipv6/route.c b/net/ipv6/route.c
>>> index f54f426..9da77e9 100644
>>> --- a/net/ipv6/route.c
>>> +++ b/net/ipv6/route.c
>>> @@ -2169,10 +2169,13 @@ int ip6_del_rt(struct rt6_info *rt)
>>> static int __ip6_del_rt_siblings(struct rt6_info *rt, struct fib6_config
>>> *cfg)
>>> {
>>> struct nl_info *info = &cfg->fc_nlinfo;
>>> + struct net *net = info->nl_net;
>>> struct sk_buff *skb = NULL;
>>> struct fib6_table *table;
>>> int err;
>>>
>>> + if (rt == net->ipv6.ip6_null_entry)
>>> + return -ENOENT;
>>
>> It looks the caller did a dst_hold(&rt->dst);
>>
>> So this new error path would leave a refcount leak.
>>
>> Note that I was not able to trigger the crash on old kernels, so it
>> would be nice to get a precise idea of bug origin.
>>
>
> Cong: do you want to send a v2 catching the null entry in ip6_route_del
> before the refcnt?
Yeah, actually it is introduced by my patch because there is already
an ip6_rt_put() in __ip6_del_rt_siblings(). So v2 is coming...
>
> for (rt = fn->leaf; rt; rt = rt->dst.rt6_next) {
> + /* do not allow deletion of the null route */
> + if (rt == net->ipv6.ip6_null_entry)
> + continue;
>
> Fixes: 0ae8133586ad net: ipv6: Allow shorthand delete of all nexthops in
> multipath route
Note, I moved the check into __ip6_del_rt_siblings() because __ip6_del_rt()
has a same check.
